KAIZENRW MANUSCRIPT ENHANCEMENT PROCESS v2.4.3 AUDIENCE NOTE: This specification is consumed primarily by orchestrating agents via prompt context, which is why curly-brace symbol notation is used throughout. Humans reviewing this document should consult the ABBREVIATION INDEX (end of file) liberally. The PATH VARIABLES REFERENCE and AGENT REFERENCE tables are the fastest entry points for understanding how artifacts and responsibilities flow through the pipeline. STATUS: This spec is fully implemented as a Claude Code plugin in the private repo github.com/RobThePCGuy/KaizenRW. End-to-end runs exist; the pipeline was tested against the author's own novel-in-progress. A patent application covering the underlying approach was filed and is not being pursued. The author has settled the question the pipeline was built to answer (see LIMITATIONS) and is publishing this spec as the durable artifact of that work. The implementation repo remains private. LIMITATIONS: AI surpasses humans in many ways. Writing is not one of them. The pipeline works. End-to-end runs against the author's manuscript produced clean, polished prose. The output is good in the way an article is good, not in the way a novel is good. It is too polished to feel like the author's. That is the actual problem, not "AI fails to generate broken sentences." AI succeeds at generation, well enough that the result stops being recognizable as the author's. A pipeline that rewrites prose normalizes prose, and normalized prose is no longer the writer's, regardless of how well-tuned the voice fingerprint or how rigorous the supervisor cycle. This pipeline cannot solve that. The architecture is the symptom, not the cure. The patterns inside (schema versioning, foundation-lock hashes, cycle-tiered thresholds, constraint registry plus mechanical-sign verification, escalation packets) hold up in domains where AI generation is appropriate. Manuscript enhancement, in the author's lived experience after building and running this end-to-end, is not one of them. NUMERIC THRESHOLDS: All numeric values in this spec (cycle thresholds, sensory density bands, voice marker floors, supervisor scores, retry caps, byte-diff percentages) are placeholder heuristics, not empirically tuned. They were not derived from a held-out evaluation suite. Replace per your manuscript, your domain, and whatever evals you actually run. PRIOR ART: This spec is framework-agnostic by design. LangGraph, CrewAI, AutoGen, and DSPy implement most of the multi-agent state-machine primitives this design assumes. A real implementation should build on one of them rather than re-derive checkpointing, role routing, supervisor loops, or compile-time metric tuning from scratch. COST, EVALS, LATENCY: Cost is not modeled in this spec. A 9-phase, 21-agent pipeline calling Sonnet-tier models would conservatively run $15-30 per chapter, more with Opus. A 30-chapter novel is in the $500-1000 range per full pass. Evaluation suite is not implemented; there is no held-out manuscript regression test. Per-chapter latency at full pipeline depth (4 passes plus idempotency reruns plus supervisor plus vision plus ripple analysis) is estimated at 2 to 4 hours, not measured. Real authors will not wait that long; a real implementation needs a streaming/partial-result story this spec does not specify. ================================================================================ PHASE 0: PROJECT INITIALIZATION ================================================================================ 1. A User {U} uploads a Manuscript {M} and initiates the project. a. {U} provides the title, author name, and Authorial Requirements {REQ} to steer enhancements. b. {U} optionally specifies if {M} is part of a Series {SER} with a book number. c. {U} specifies Voice Source {VSRC} for voice modeling: 1. user_sample: {U} provides 500-1000 words of their own writing (recommended for AI-drafted manuscripts) 2. author_mirror: {U} names a published author to emulate 3. blend: {U} provides sample + mirror author with blend weights 4. manuscript: Extract voice from {M} itself (only for human-written manuscripts) d. {U} optionally provides User Voice Sample {UVS} (500-1000 words of their own prose): 1. Recommended: Include two registers (dialogue scene + introspection scene) 2. This becomes the authoritative voice signal, separate from {M} e. {U} optionally provides gold excerpt samples for additional voice conditioning. 2. The Initializer {INIT}: a. Determines the Manuscript Source {MSRC} (explicit > heuristic): 1. Preferred: {U} selects the manuscript file explicitly (upload selection or path). 2. If provided: loads Manuscript Manifest {MMAN} (manifest.json) with manuscript_path, format, encoding. 3. Fallback: heuristic detection (files >10KB, excluding README/LICENSE) BUT requires {U} confirmation. 4. Records selection to status/manuscript-selection.json (hash, bytes, detected format). b. Creates the Project Directory Structure {PROJ}: 1. chapters/{original, revised, polished} 1.1 chapters/revised/.working (RW intermediate prose drafts) 2. kb/{scene-states} 3. avd/{gold-excerpts, voice-sample} 4. meo/ 5. plans/ 6. reviews/{qa, supervisor, vision, series} 7. updates/{retroactive-fixes} 8. diagnostics/ 9. metrics/ 10. polish-reports/ 11. status/ 12. versions/ 13. schemas/ 14. overrides/ 15. escalations/ c. Extracts {M} into individual Manuscript Chapters {MC} using chapter header patterns: 1. Prologue → chapter-00.md 2. Chapter N → chapter-NN.md 3. Interlude → chapter-NNa.md (multiple interludes between same chapters: NNa, NNb, NNc, ...) 4. Epilogue → chapter-[LAST+1].md 5. Writes Chapter Map {CMAP}: chapter_id, title, start/end offsets, wordcount. 6. Runs Chapter Map Validator {CMV} (fast sanity checks): a. Order monotonic, no duplicate IDs, no missing numbers (unless explicitly labeled interlude). b. No empty chapters; flags extreme wordcount outliers for review. c. Total extracted words reconcile to {M} within tolerance (default: >= 98%). 7. If {CMV} FAIL: requires Manual Chapter Map Override {CMAPO} from {U}: a. Stored to overrides/chapter-map.override.json b. {INIT} re-extracts using the override map. d. Creates template files: authorial-requirements.md, project-status.md. e. If {UVS} provided: 1. Stores to avd/voice-sample/user-voice.md 2. If two registers provided: splits to avd/voice-sample/register-dialogue.md and avd/voice-sample/register-introspection.md f. If {SER} specified: 1. Creates Series Link {SLINK} configuration file. 2. If Book 1: Creates empty Series Canon {SCAN} structure. 3. If Book 2+: Inherits from existing {SCAN}; marks content [INHERITED FROM SERIES]. g. Outputs: {PROJ}, {MC} files, {UVS} (if provided), {SLINK} (if {SER}). h. Initializes Schema Registry {ASCH}: 1. Writes JSON schema files to schemas/ (meoj, kb, ssf, creg, ssnap, stat, override, escalation, change-manifest, supervisor-metrics, vision-metrics, polish-metrics, chapter-map, calibration, theme-keywords, arc-ledger, rules, propagation, version-record, book-registry, checkpoint, manuscript-selection, synthetic-origin-likelihood, voice-drift-analysis, kb-merge-log, delta-merge-queue, bfa-log, cpa-log, approval-manifest). 2. Requires schema_version in every JSON artifact. 3. All agents MUST validate inputs against schemas/ before reading/writing. i. Initializes Foundation Lock {LOCK}: 1. Creates versions/foundation-lock.json with schema_version, created_at, and nullable hash fields: a. avd_hash: null b. kb_hash: null c. cal_hash: null d. scan_hash: null 2. Producer agents update {LOCK} when their artifact is created or intentionally regenerated. 3. {LOCK} is the single source of truth for foundation artifact pins. 4. Lock reconciliation rule: if a non-null lock hash mismatches the on-disk artifact, HALT and surface to {U}. If a lock hash is null but the artifact exists on disk, the producer agent failed to update {LOCK}; run reconciliation (recompute hash, write to {LOCK}, log to status/lock-reconcile.json) and HALT until {U} confirms scope of dependent regeneration. 5. Resume contract on reconciliation halt: {STAT}.lock_state = "halted_reconcile" gates re-entry. No agent in any phase MAY proceed against this chapter until {STAT}.lock_state returns to "clear". On {U} confirmation, the orchestrator writes lock_state = "clear" and emits a resume_log entry to status/lock-reconcile.json. If {U} walks away mid-halt, checkpoint files persist indefinitely; the chapter is excluded from {FR} discovery (Step 23b) until lock_state resolves. 6. Hash-pinning contract: {LOCK} guarantees all agents in a single cycle read identical input bytes for {AVD}/{KB}/{CAL}/{SCAN}. It does NOT guarantee deterministic output (the runtime is LLM-nondeterministic). Use the lock for cache invalidation and audit-trail integrity, not for reproducibility. ================================================================================ PHASE 1: DIAGNOSTIC ANALYSIS ================================================================================ 3. The Diagnostic Analyzer {DA}: a. Inputs {MC} (all chapters). b. Analyzes four dimensions per chapter: 1. Pacing Analysis: a. Segments chapter into 8-15 units b. Scores each for pacing (0-100), tension (0-100) c. Identifies type: action/dialogue/description/introspection/transition d. Calculates curves: rising/falling/wave/flat/climactic/front_loaded/flat_middle 2. Sensory Analysis: a. Target density: 1 per 150-200 words b. Maps five senses distribution c. Ideal balance: sight 30-40%, sound 20-30%, touch 15-25%, smell 10-15%, taste 5-10% d. Detects: clustering, repetition, deserts (500+ words without), stacking 3. Emotional Analysis: a. 10 segments with intensity (0-100), valence (-100 to +100) b. Identifies arc type: rising/falling/wave/climactic/valley/flat/zigzag/bookend c. Locates peak, low point, turning points, flat zones 4. Foreshadowing Analysis: a. Identifies setups: chekhov_gun, prophecy, hint, warning, symbolic, question b. Identifies payoffs: resolution, callback, revelation, subversion, partial c. Detects repetitive foreshadowing (BLOCKER) c. Outputs Chapter Diagnostic {CDIAG} per chapter with priority areas scored: 1. 0-40: CRITICAL 2. 41-60: HIGH 3. 61-75: MEDIUM 4. 76-90: LOW 5. 91-100: NONE d. Outputs Manuscript Diagnostic {MDIAG} with aggregate scores and weakest/strongest areas. e. Outputs Foreshadowing Registry {FREG} (manuscript-level): 1. Normalizes each setup (type, anchor, chapter_id, scene_id if available). 2. Computes setup fingerprints (for repetition detection across chapters). 3. Writes to kb/foreshadowing-registry.json and metrics/foreshadowing-registry.md. f. Computes Synthetic Origin Likelihood {SOL} from {MC}: 1. Scores manuscript prose against synthetic-prose markers: sentence-rhythm uniformity, generic transition phrases, low signature-word density, repeated structural templates. 2. Output: {SOL} score 0.0-1.0 with per-marker contributions. 3. Writes to diagnostics/synthetic-origin-likelihood.json. 4. Advisory only; consumed by {CALR} (Step 4) to flag voice_source mismatch risk. ================================================================================ PHASE 2: CALIBRATION ================================================================================ 4. The Calibrator {CALR}: a. Inputs {MDIAG} (if exists), {VSRC}, and {U} responses to M.A.P. (Manuscript Author Profiling) game. b. M.A.P. Scenario Questions (7 core): 1. Exposition Dilemma: How to handle skipped world-building 2. Beloved Darling: Scene without plot purpose 3. Voice Drift: AI revision smoother but different 4. Structural Gamble: Reordering scenes for pacing 5. Length Dilemma: Manuscript exceeds industry standard 6. Perfectionist's Trap: 92% quality - loop or accept? 7. AI Partnership: Tool vs editor vs partner vs amplifier c. M.A.P. Microcalibration (2-3 A/B choices): 1. Presents same beat from {M} written in two different styles 2. {U} picks preferred version (A or B) 3. Provides cleaner signal for: rhythm, sentence density, exposition tolerance 4. If {MDIAG} exists: uses actual manuscript situations for A/B pairs d. If {MDIAG} exists: adapts core scenarios to actual manuscript issues. e. Derives enhancement parameters: 1. strategic_intent: refine | rework | reimagine 2. length_bias: compress | maintain | expand a. Target bands (used by {RW} Pass 4 success tests): - compress: -10% to -25% wordcount delta vs {MC} - maintain: ±5% wordcount delta vs {MC} - expand: +10% to +25% wordcount delta vs {MC} 3. voice_threshold: 0.80-0.98 4. continuity_thresholds: {cycle_1: 92, cycle_2: 90, cycle_3: 88} 5. continuity_floor: 80 (hard minimum; below this triggers mandatory escalation regardless of cycle) 6. supervisor_thresholds: {cycle_1: 95, cycle_2: 88, cycle_3: 81} 7. supervisor_floor: 70 (hard minimum; below this triggers mandatory escalation regardless of cycle, no Rescue retry) 8. output_control: {verbosity: minimal | standard | verbose} 9. trust_level: conservative | moderate | aggressive 10. darling_tolerance: low | medium | high (consumed by {RW} Pass 1-2 when removing scenes flagged as darlings) 11. exposition_tolerance: low | medium | high (consumed by {RW} Pass 3 grounding decisions and {SUP} blocker check 8) 12. pacing_priority: action | balanced | atmosphere (consumed by {CTP} Pass 1 prioritization) 13. voice_source: user_sample | author_mirror | blend | manuscript (from {VSRC}) 14. voice_blend_weights: {user: 0.0-1.0, mirror: 0.0-1.0} (if blend) 15. mirror_author: author name (if author_mirror or blend) 16. gee_supplement_opt_in: per-scene-type list (default empty; controls {GEE} fallback to {MC} when voice_source != manuscript) f. Voice source mismatch check (uses {SOL} from Step 3f): 1. If voice_source == manuscript AND {SOL} >= 0.65: WARN to {U} with synthetic markers and recommend voice_source = user_sample. 2. {U} may proceed (recorded as override in {CAL}.voice_source_override = true) or change voice_source. g. M.A.P. skip handling: 1. If {U} skips M.A.P., applies Default Standard Profile: strategic_intent=refine, length_bias=maintain, voice_threshold=0.92, continuity 92/90/88 with floor 80, supervisor 95/88/81 with floor 70, verbosity=minimal, trust=moderate, darling=medium, exposition=medium, pacing=balanced, voice_source from {VSRC} only. 2. Records {CAL}.map_skipped = true. h. Outputs Calibration Profile {CAL}. i. Writes version record to versions/calibration-version.json (content_hash, schema_version, created_at). j. Updates versions/foundation-lock.json with cal_hash. ================================================================================ PHASE 3: FOUNDATION BUILDING ================================================================================ 5. The Manuscript Visionary {MV}: a. Inputs {MC} (all chapters), {REQ}, {CAL} (if exists), {UVS} (if provided). b. Can run in parallel with Knowledge Base Builder. c. Extracts Voice Fingerprint {VFP} based on {CAL}.voice_source: 1. If user_sample: Compute Target {VFP} from {UVS} a. If registers provided: extract separate fingerprints per register (dialogue/introspection) 2. If author_mirror: Compute Target {VFP} from {CAL}.mirror_author's known style patterns 3. If blend: Compute Target {VFP} blending {UVS} and mirror author per {CAL}.voice_blend_weights 4. If manuscript: Extract {VFP} from {MC} (traditional mode - only for human-written manuscripts) d. {VFP} Metrics (computed from voice source, not {M}): 1. Sentence length: mean ± std dev (e.g., 14.2 ± 6.8 words) 2. Structure mix: Simple/Compound/Complex % (e.g., 45%/30%/25%) 3. Vocabulary tier: grade level (Flesch-Kincaid) 4. Fragment rate: per 1000 words 5. Dialogue ratio: % of total text 6. Signature words: top 5-10 distinctive terms Note: {VFP} metrics are inputs to retrieval and conditioning, not a definition of voice. They cannot capture rhythm across a whole novel, intent, restraint, or the writer's argument with the text. They are the histogram beneath the prose, not the prose. Treat downstream voice scoring as a leading indicator of drift, not as ground truth on whether the prose is the writer's. e. If voice_source != manuscript: Optionally computes Manuscript Voice Fingerprint {MVFP}: 1. Extracts same metrics from {MC} 2. Compares {MVFP} to Target {VFP} 3. Calculates AI Drift Score: how "synthetic" the current draft appears 4. Logs drift report to diagnostics/voice-drift-analysis.json f. Identifies Thematic Core: central themes, emotional register, narrative purpose. g. Maps Character Development: arcs, key moments, transformations for major characters. h. Documents Discipline Patterns: 1. Foreshadowing style: telegraph vs trust reader 2. Exposition style: dialogue dumps vs discovery through action i. Establishes 5 measurable, objective, falsifiable Success Criteria. j. Outputs: 1. Authorial Vision Document {AVD}: 8 sections including {VFP} 2. Theme Keywords {TKW}: weighted keyword lists with baseline densities 3. Character Arc Ledger {ARC}: expected states per chapter per character 4. Voice Drift Analysis {VDA} (if voice_source != manuscript): consumed by {RW} Pass 4 (Step 13d.5) to prioritize voice-restoration targets; consumed by {SUP} (Step 17e.3) as evidence input for Voice Integrity scoring. 5. Writes version record to versions/avd-version.json (content_hash, schema_version, created_at) 6. Updates versions/foundation-lock.json with avd_hash. 6. The Knowledge Base Builder {KBB}: a. Inputs {MC} (all chapters). b. Can run in parallel with {MV}. c. Extracts and organizes 9 sections: 1. Character Registry: names, traits, roles, aliases, arc, first_appearance 2. Location Catalog: settings, descriptions, significance 3. Timeline Framework: events, sequences, causality 4. Object/Symbol Inventory: items, motifs, symbolic meaning 5. Relationship Map: character connections, dynamics, evolution 6. Plot Thread Tracker: active threads, resolution status 7. Contradiction Log: detected conflicts with severity (CRITICAL/MAJOR/MINOR) 8. Name Normalization: canonical names, aliases, variants 9. Cross-Reference Index: element→chapters, chapter→elements d. Outputs Knowledge Base {KB}. e. Writes version record to versions/kb-version.json (content_hash, schema_version, created_at). f. Updates versions/foundation-lock.json with kb_hash. 7. The Gold Excerpt Extractor {GEE} (conditional): a. Inputs {MC}, {AVD}, {CAL}, {UVS} (if exists). b. Source Selection based on {CAL}.voice_source: 1. If user_sample or blend: {UVS} is PRIMARY source for {GOLD} a. User's 500-1000 words becomes avd/gold-excerpts/user/*.md b. If registers provided: maps to scene types (dialogue, introspection) c. {MC} supplement for missing scene types: OFF by default when voice_source != manuscript. Requires explicit per-scene-type opt-in via {CAL}.gee_supplement_opt_in. When enabled, supplemented excerpts are tagged [FROM MANUSCRIPT] and given lower weight in voice conditioning. 2. If author_mirror: Skips extraction (voice comes from mirror author patterns) 3. If manuscript: Extracts from {MC} as traditional mode (only if human-written) c. Scene Types for extraction (if needed): 1. action: physical conflict, chase, urgency 2. introspection: alone, thinking, memories 3. dialogue: conversation-focused 4. description: setting, atmosphere 5. tension: threat, stakes rising 6. emotional: grief, joy, revelation 7. transition: time skip, scene bridge d. Identifies exemplary passages (200-400 words) for each type. e. Extracts 2-3 passages per type with voice markers noted. f. Priority Order: 1. avd/gold-excerpts/user/ (from {UVS}) - highest authority 2. avd/gold-excerpts/curated/ (if {U} provided additional excerpts at Step 1e) 3. avd/gold-excerpts/manuscript/ (from {MC}, only if voice_source == manuscript) g. Outputs Gold Excerpts {GOLD} with source tagging. 8. The Literary Master Planner {LMP}: a. Inputs {MC}, {AVD}, {KB}, {REQ}, {CDIAG} (if exists), and {CAL}. 1. Loads versions/avd-version.json, versions/kb-version.json, versions/calibration-version.json, and versions/foundation-lock.json. 2. If {SLINK} exists, also loads versions/series-canon-version.json. b. Analyzes each chapter against: {AVD} success criteria, {KB} continuity requirements, narrative arc position, cross-chapter dependencies, diagnostic priorities. c. For each chapter documents: 1. Current Issues: specific weaknesses, missed opportunities, continuity gaps 2. Enhancement Goals: measurable objectives 3. Actionable Suggestions: specific changes with priority and pass affinity 4. Dependencies: which chapters must be enhanced first (Plot/Character/Information) 5. Rationale: how goals connect to {AVD} success criteria d. Assigns priorities: 1. P1 (Critical): blocks approval 2. P2 (Important): this cycle 3. P3 (Nice-to-have): if time permits e. Tags pass affinity: Story | Emotion | Clarity | Polish. f. Outputs in both formats: 1. MEO JSON {MEOJ}: JSON format (primary, enables dynamic updates) 2. Master Enhancement Outline {MEO}: Markdown format (human-readable) g. Pinning rule: {MEOJ} MUST include pinned_input_hashes from {LOCK}: avd_hash, kb_hash, cal_hash, scan_hash (if series). 9. The World Rules Generator {WRG} (optional): a. Inputs {KB}, {MC}. b. Identifies implicit world-building rules: 1. Magic system costs and limitations 2. Physical world constraints 3. Social/cultural norms 4. Biological traits 5. Technology limitations c. For each rule extracts: 1. Trigger patterns (keywords that activate rule) 2. Consequence patterns (what must follow) 3. Token window (how far to look) 4. Severity (CRITICAL/WARNING) 5. Exceptions d. Outputs World Rules {RULES}. ================================================================================ PHASE 4: PRE-ENHANCEMENT VALIDATION (SERIES ONLY) ================================================================================ 10. The Series Continuity Checker {SCC}: a. Only runs if {SLINK} exists. For Book 1, runs against the empty {SCAN} created at Step 2f.2.b and trivially passes (no canon to violate). Continues to write SCC artifacts for audit trail. b. Inputs {MC}, {SLINK}. c. Loads from {SCAN} (and {OVR} if exists for chapter): 1. Series Knowledge Base {SKB} 2. Series Timeline {STL} 3. Series Voice Baseline {SAVD} 4. locked-elements.json d. Validates original chapters against {SCAN}: 1. Character facts (deceased status, traits, relationships) 2. Timeline events (sequences, dates) 3. World rules (magic/tech, geography) 4. Locked elements (any modification = CRITICAL) e. Applies inheritance mode policy from {SLINK} (single decision model): 1. Classifies findings into: CRITICAL | MAJOR | MINOR (classification rules vary by mode). 2. Locked elements: any modification = CRITICAL and ALWAYS blocks (no override). 3. If an {OVR} exists, SCC may downgrade MAJOR/MINOR findings within scope, but never locked-element CRITICALs. 4. Decision: if ANY CRITICAL exists → {FAIL} (blocker). 5. Score model (if no CRITICAL): a. Start score = 100 b. Strict mode penalties: MAJOR = -10, MINOR = -2, pass if score >= 95 c. Permissive mode penalties: MAJOR = -5, MINOR = -1, pass if score >= 85 f. Decision: 1. {PASS}: Proceed to enhancement 2. {FAIL}: Address CRITICAL violations before proceeding ================================================================================ PHASE 5: PER-CHAPTER ENHANCEMENT LOOP ================================================================================ START [PER-CHAPTER ENHANCEMENT LOOP] {U} is presented with chapter processing interface. Enhancement proceeds in dependency order per {MEO}. 11. The Scene Parser {SP} (Pre-Enhancement): a. Inputs {MC} for target chapter. b. Identifies scene boundaries: location changes, time jumps, POV shifts, scene breaks. 1. Labels each boundary with boundary_type: continuous | time_jump | location_jump | pov_shift | hard_cut | montage | unspecified c. For each scene extracts State Object at START and END: 1. time_of_day: morning/afternoon/evening/night/dawn/dusk/unspecified 2. time_relative: same_day/next_day/days_later/weeks_later/unspecified 3. weather: clear/rain/snow/storm/fog/unspecified 4. location: location name 5. characters_present: [list] 6. character_conditions: {health, emotional, inventory} 7. boundary_type_from_prev: continuous | time_jump | location_jump | pov_shift | hard_cut | montage | unspecified 8. NOTE on unknowns: a. 'unspecified' means UNKNOWN (not a contradiction). b. Continuity checks only compare fields when BOTH sides are specified. c. If a bridge is required, flag as MISSING_EVIDENCE (not CONTRADICTION) unless the text asserts incompatibility. d. Logs state_changes within each scene. e. Validates continuity compatibility between scenes: 1. If Scene N+1.boundary_type_from_prev == continuous: a. Scene N end_state should match Scene N+1 start_state for location, characters_present, and carried inventory b. Time/weather changes must be explicitly bridged (or flagged) 2. Else (non-continuous boundary): allows state changes, but flags internal contradictions as CONTINUITY_GAP (WARNING) f. Outputs Scene State File {SSF} (original baseline). g. Writes baseline scene-state version record to versions/chapter-[N]/ssf-baseline-version.json (content_hash, schema_version, created_at). 11.5 The Status Snapshot Builder (runRewrite context hardening): a. Purpose: produce a single, authoritative "Status Snapshot" for the chapter so downstream agents do NOT rely on stale/baseline context. b. Inputs (must be MOST RECENT versions on disk): {KB}, {SSF} (for target chapter), {STAT} (if exists), {OVR} (if exists), and versions/foundation-lock.json. c. Snapshot selection rules: 1. Prefer the latest post-rewrite {SSF} (from Step 15) if it exists for the current cycle; otherwise use the baseline {SSF} from Step 11. 2. Prefer the current master {KB} (plus any merged deltas) over any embedded "baseline" excerpts in older plans/reviews. 3. If foundation hashes in versions/foundation-lock.json do not match the loaded {KB}/{CAL}/{AVD}/{SCAN if series} inputs: HALT and force regeneration of plans (prevents drift). 4. {SSF} is chapter-local and is NOT stored in {LOCK}; compute ssf_hash directly from the selected {SSF} and compare it only against the ssf_hash pinned in {SSNAP}/{CTPL}. d. Outputs Status Snapshot {SSNAP} (JSON; schema_version required): 1. chapter_id, cycle, created_at 2. pinned_input_hashes: avd_hash, kb_hash, cal_hash, ssf_hash, scan_hash (if series) 3. per_scene: scene_id → {location, boundary_type_from_prev, characters_present_start, characters_present_end} 4. global_presence_audit: character → last_known_location (or "unknown"), on_page (bool), confidence 5. active_mechanical_expectations: world-events in play + expected sensory signs (from {RULES} if exists; else empty) 6. overrides_in_effect: subset of {OVR} that affects continuity/rules/abilities e. Output path: narrative-project/status/chapter-[N]-snapshot.json f. Consumption rule: {CTP} and {RW} MUST treat {SSNAP} as higher priority than any older/baseline chapter state. g. Hash provenance rules: 1. ssf_hash is computed directly from the selected {SSF}. 2. avd_hash, kb_hash, cal_hash, scan_hash are copied from {LOCK}. 3. If a non-null {LOCK} hash mismatches its on-disk artifact: HALT and trigger plan regeneration per Step 11.5.h. 4. If a {LOCK} hash is null but the artifact exists on disk: invoke lock reconciliation (Step 2i.4); HALT until {U} confirms regeneration scope. h. Plan invalidation cascade: 1. Foundation hash change ({AVD}, {KB}, {CAL}, {SCAN}) invalidates {MEOJ} and ALL downstream {CTPL} files. Regenerate per Concurrency & Plan Invalidation section. 2. Chapter-local ssf_hash change invalidates only the affected chapter's {CTPL}. 12. The Chapter Tactical Planner {CTP}: a. Determines current cycle (1, 2, or 3): 1. If {STAT} exists for chapter: cycle = {STAT}.current_cycle + 1 2. Else: cycle = count(plans/chapter-[N]-ctp-cycle*.md) + 1 3. Hard cap: max cycle = 3 4. {CTP} writes the new cycle to {STAT}.current_cycle as its FIRST action. {STAT}.current_cycle is the single source of truth for the in-flight cycle; downstream agents ({SUP}, {CC}, {VFA}) MUST read from {STAT}, not infer from filenames. b. Inputs {MC}, {MEO}, {AVD}, {KB}, {CAL}, {CDIAG} (if exists), previous tactical plan (if cycle 2+). c. Mode Selection based on {CAL}.strategic_intent and cycle: 1. Standard Mode (cycles 1-2): Full {MEO} coverage per strategic_intent 2. Rescue Mode (cycle 3): P1 goals only, 70-80% preservation, voice over enhancement d. Completes Preflight Phase: 1. Plot Facts {PF}: events that MUST occur, pinned to scene locations 2. Emotional Non-Negotiables {EN}: feelings that MUST be evoked, anchored to moments 3. Motifs & Mechanics {MM}: symbols with frequency rules (2-3 times per arc) 4. Constraint Registry {CREG} (MANDATORY; for EVERY scene_id in {SSF}): a. Presence Audit: explicit map of "who is where" (prevents any character being in two places at once). - Must list ON-PAGE vs OFF-PAGE characters with last-known location (or "unknown"). b. Ability Lock: explicit negatives for perception/action constraints (e.g., "Ryan cannot see electrical sparks"). - Derived from {KB} traits/limitations, scene conditions, {RULES} (if exists), and {OVR} (if exists). c. Mechanical Anchors: mandatory sensory/worldbuilding cues tied to world-events relevant to the scene. - Each anchor includes trigger, required sign(s), window (e.g., within 2 paragraphs), and severity (CRITICAL/WARNING). d. Output requirements: - Must appear in {CTPL} as a per-scene registry table. - Must be serialized to narrative-project/kb/scene-states/chapter-[N]-constraints.json with schema_version. e. Extracts Voice Marker Floors from {AVD}: 1. Counts POV character markers in {MC} 2. Sets floor = MAX(original_count - 1, 2) f. Answers strategic questions: 1. What is chapter's job in the manuscript? 2. What should reader feel at end? 3. If only ONE change, what would it be? g. Creates 4-Pass Structure with step counts per strategic_intent: | Intent | Pass 1-2 | Pass 3 | Pass 4 | Preservation | | refine | 1-2 steps | 1-2 | 1 | 80-90% | | rework | 2-3 steps | 2-3 | 1-2 | 50-70% | | reimagine | 3-4 steps | 2-3 | 2-3 | 30-50% | | rescue | 1-2 steps | 1-2 | 1 | 70-80% | h. Pass Definitions: 1. Pass 1 - Story Structure: beats, pacing, stakes, GOO (goal-obstacle-outcome) 2. Pass 2 - Character & Emotion: emotional shifts, relationships, showing vs telling 3. Pass 3 - Clarity & Setting: grounding, sensory detail, worldbuilding 4. Pass 4 - Line-Level Polish: verb strength, dead weight, rhythm, voice i. Each step specifies: Location, Action, Success Test (falsifiable), KB Check, and Constraint Checks (scene_id + {CREG} references). j. Adds quality gate as final step of each pass (60-second verification). k. Outputs Chapter Tactical Plan {CTPL} with header: 1. Mode (Standard/Rescue) 2. Strategic Intent 3. Length Bias 4. Target Preservation % 5. Diagnostic Summary (if {CDIAG} exists) 6. Voice Marker Floors 7. Pinned Hashes (from versions/foundation-lock.json): avd_hash, kb_hash, cal_hash, scan_hash (if series) 8. Status Snapshot: {SSNAP} path + ssnap_hash 9. Constraint Registry: {CREG} path + creg_hash (all scenes covered) 13. The Chapter Rewriter {RW}: a. Inputs {MC}, {CTPL}, {KB}, {AVD}, {GOLD} (if exists), {SSF}, {CREG}, {SSNAP}. b. If Revised Chapter {REV} exists from prior attempt, uses as base; otherwise uses {MC}. c. Validates inputs per Input Validation Protocol (files exist, >500 bytes). d. Voice Conditioning (if {GOLD} exists): 1. Classifies current section by scene type 2. Loads matching excerpts (primary: full text excerpt, secondary: 150-word excerpt for context-budget cases) 3. Constructs conditioning block with techniques to emulate 4. Prepends to execution context 5. If voice_source != manuscript AND {VDA} exists: loads top voice-drift offenders from {VDA} as Pass 4 priority targets (e.g., generic transition phrases, sentence-rhythm uniformity hotspots). e. For each {CTPL} step in order: 1. Runs idempotency check: if success test passes, skip step and record entry in {CMAN} with status='skipped_idempotent', success_test_result='pass'. If ALL CTPL steps skip, RW writes prose unchanged AND emits {CMAN} with all_steps_skipped=true; {OV} accepts this as Idempotent No-Op (Step 14d.1.a). 2. Consults {KB} sections identified in step before modifying 3. Executes ONLY what step specifies - no interpretation 4. Matches {AVD} fingerprint for new/modified prose 5. Adds minimal bridging if needed (max 1-2 sentences), including required Vanishing Character beats at scene transitions (see Vanishing Character Protocol, Step 13h.5) 6. Verifies no continuity violations against {KB} and {SSF} 6.1 Verifies Constraint Registry compliance for the affected scene(s): - Presence Audit: no character is asserted in a location that contradicts {CREG} without an on-page transition beat. - Ability Lock: no prohibited perception/knowledge claims are attributed to locked characters. - Mechanical Anchors: if a trigger is present/near, required sensory sign(s) are included. 7. Documents new facts in KB Delta {DELTA} only (not master {KB}) 8. Runs step's success test before proceeding f. After each trimming step: counts voice markers. 1. If below floor: restores removed instances 2. If still below floor after restoration: FAIL g. Layer Separation Rule: each pass focuses ONLY on its layer. h. Guardrails enforced: 1. No {PF} re-establishing (trust reader memory) 2. No {EN} replanting (plant anticipation once per arc) 3. {MM} frequency control (2-3 times per arc) 4. No lecture-mode dialogue (max 2-3 sentences before action/reaction) 5. Vanishing Character Protocol: if a character is present in Scene S but absent in Scene S+1, write a narrative beat explaining the exit/transition. 6. Ability Lock Enforcement: locked characters cannot perceive/know prohibited signals; route perception through allowed characters/senses. 7. Mechanical Anchors: when a world-event trigger is in play, include the established sensory sign(s) (per {CREG}/{RULES}). i. Outputs two distinct artifacts: 1. Draft Prose {D} → chapters/revised/.working/chapter-[N]-cycle-[C]-draft.md (prose only, no agent commentary, no manifest, no delimiters). 2. Change Manifest {CMAN} → metrics/chapter-[N]-change-manifest-cycle-[C].json (validates against schemas/change-manifest.schema.json). 3. Manifest entries (one per CTPL step, executed or skipped): a. step_id (from {CTPL}) b. status: executed | skipped_idempotent | failed c. anchor (scene_id from {SSF} preferred; else short paragraph anchor quote) d. intent (what changed and why; for skipped, why the success test already passed) e. success_test_result (pass/fail + 1-line note) 4. Manifest top-level fields: chapter_id, cycle, all_steps_skipped (bool), entries[]. 5. {D} contains ONLY prose. Any manifest delimiter or agent commentary in {D} is a defect caught by {VFA} QA Lint (Step 18h.1.a.1). 14. The Output Validator {OV}: a. Inputs {MC}, {D}, {CMAN}, and {CTPL} (preferred). b. Computes change signals (against {D} prose only; {D} contains no manifest by construction): 1. exact_match: SHA-256({MC}) == SHA-256({D}) 2. byte_diff_percent = ABS(bytes({D}) - bytes({MC})) / bytes({MC}) 3. changed_sentence_count (sentence-level diff) 4. changed_paragraph_count (paragraph-level diff) c. Computes plan/semantic signals (from {CMAN}): 1. has_change_manifest: {CMAN} exists and validates against schemas/change-manifest.schema.json 2. manifest_step_coverage: count of CTPL step_ids referenced in {CMAN} with anchors (scene_id or paragraph anchor) AND status='executed' 3. has_success_test_results: {CMAN} includes at least 1 success-test result (pass/fail + note) for an executed step 4. all_steps_skipped: {CMAN}.all_steps_skipped == true (every step's idempotency check passed on entry) d. Decision: 1. exact_match: a. If has_change_manifest AND all_steps_skipped: {PASS} as Idempotent No-Op (chapter requires no work this cycle; {STAT} marks cycle complete with reason='idempotent'). b. Else: {REJECT} as Silent Write Failure → returns to {RW} with "WRITE_FAILED: Output identical to input AND no idempotency manifest". Max 2 retries, then {ESCALATE}. 2. If byte_diff_percent >= 10%: {PASS} to Scene Parser 3. If byte_diff_percent >= 1% AND (changed_sentence_count >= 3 OR changed_paragraph_count >= 1): {PASS} to Scene Parser 4. Else if has_change_manifest AND manifest_step_coverage >= 1 AND has_success_test_results: {PASS} to Scene Parser 5. Else: {REJECT} as Insufficient/Unverifiable Delta a. Returns to {RW} with reason code: - WRITE_TOO_SMALL (delta below minimum) - WRITE_NO_MANIFEST (missing or invalid {CMAN}) - WRITE_MANIFEST_INVALID (no anchors / no success tests) b. Max 2 retries, then {ESCALATE} 15. The Scene Parser {SP} (Post-Rewrite): a. Inputs {D} prose only. MUST NOT read {CMAN}. b. Updates {SSF} for revised chapter. c. Validates scene-level continuity compatibility (using {SSF} boundary types): 1. continuous boundaries: end_state matches next start_state for location, characters_present, and carried inventory (unknown fields ignored) 2. non-continuous boundaries: allow jumps; flag CONTRADICTION only when both sides are specified and incompatible; otherwise flag MISSING_EVIDENCE when a bridge is implied but not shown 3. Time/weather transitions logical d. Flags gaps as CONTINUITY_GAP with severity (ERROR/WARNING). e. Writes post-rewrite scene-state version record to versions/chapter-[N]/ssf-cycle-[C]-version.json (content_hash, schema_version, created_at). 16. The Continuity Checker {CC}: a. Inputs {D}, {KB}, {SSF}, {CREG} (if exists), {SSNAP} (if exists), {CAL}, {STAT} (if exists), {RULES} (if exists), and {OVR} (if exists). b. If {SLINK} exists: also inputs {SKB}, {STL}, locked-elements.json. c. Extracts all entity mentions from {D}: characters, locations, objects, timeline references. d. Cross-references against {KB}: 1. Character Registry: names, traits, appearances 2. Location Catalog: settings, details 3. Timeline Framework: sequences, dates 4. Object Inventory: items, properties e. Validates scene state continuity from {SSF}. f. If {RULES} exists: validates trigger/consequence pairs within token windows. f.1 Mechanical Sign Verification (MANDATORY): 1. Builds mechanical sign expectations from {CREG}.Mechanical_Anchors (preferred) and {RULES} (fallback). 2. For each expectation whose trigger is asserted or strongly implied in {D}: a. Verify at least one required sensory sign appears within the specified window. b. If missing: flag as MISSING_EVIDENCE or RULE_VIOLATION with severity based on expectation severity. 3. Validates Ability Locks from {CREG}: no prohibited perception/knowledge assertions appear in {D}. 4. Validates Presence Audit coherence across adjacent scenes: disappearance without an on-page beat → MISSING_EVIDENCE (or CRITICAL if continuous boundary). g. If {SLINK} exists: cross-references against {SCAN} files. h. Categorizes issues by issue type: 1. CONTRADICTION: explicit conflict between specified facts 2. MISSING_EVIDENCE: gap where continuity is implied but not supported on-page 3. STYLE/VOICE: voice marker drift (if enabled) 4. RULE_VIOLATION: world-rule trigger/consequence broken 5. SERIES_CANON: series-only violations 6. MECHANICAL_SIGN: missing/incorrect sensory signals tied to world-events (from {CREG}/{RULES}) i. Categorizes issues by severity: 1. CRITICAL (blocker): contradiction, timeline_error, scene_gap, rule_violation_critical, locked_element_violation 2. WARNING: inconsistency, name_variation, location_error, rule_violation_warning, scene_gap_minor 3. SUGGESTION: missing_reference (may need {KB} addition) j. Flags KB delta suggestions for new facts. k. Calculates Continuity Score {CSCORE} (0-100). l. Determines continuity threshold from {CAL} (or defaults): 1. cycle = {STAT}.current_cycle (authoritative source). 2. Fallback if {STAT} missing: parse cycle from {CTPL} filename (chapter-[N]-ctp-cycle[C].md). 3. Final fallback: cycle = count(plans/chapter-[N]-ctp-cycle*.md) (fallback: 1). 4. threshold = {CAL}.continuity_thresholds.cycle_[cycle] (default: 92). 5. floor = {CAL}.continuity_floor (default: 80). m. Decision: 1. Hard floor check: if {CSCORE} < floor, mandatory {ESCALATE} regardless of cycle (no Rescue retry path). 2. {PASS}: {CSCORE} >= threshold AND no CRITICAL issues → proceed to Supervisor. 3. {WARN}: {CSCORE} >= threshold but has warnings → proceed with notes. 4. {FAIL}: {CSCORE} < threshold OR CRITICAL present. a. Cycles 1-2: Return to {RW} with specific violations to fix. b. Cycle 3: {ESCALATE} to {U}: manual fix / add override / skip. 17. The Chapter Supervisor {SUP}: a. Determines current cycle (1, 2, or 3): 1. cycle = {STAT}.current_cycle (authoritative source). 2. Fallback if {STAT} missing: parse cycle from {CTPL} filename (chapter-[N]-ctp-cycle[C].md). 3. Final fallback: cycle = count(metrics/chapter-[N]-supervisor-cycle-*.json) + 1. 4. Hard cap: max cycle = 3. b. Inputs {MC}, {D}, {CTPL}, {MEO}, {AVD}, {KB}, {ARC} (if exists), {TKW} (if exists). c. Loads {CAL} for supervisor_thresholds and selects threshold by cycle: 1. Cycle 1: {CAL}.supervisor_thresholds.cycle_1 (default: 95%) 2. Cycle 2: {CAL}.supervisor_thresholds.cycle_2 (default: 88%) 3. Cycle 3: {CAL}.supervisor_thresholds.cycle_3 (default: 81%) d. Follows Fresh Eyes Protocol: reads {D} FIRST (forms impressions), THEN compares to {MC}, {CTPL}, {KB}. e. Scores 5 dimensions per weights: 1. MEO Alignment (25%): objectives addressed, strategic enhancements present 2. CTP Fulfillment (15%): steps completed, success tests pass 3. Voice Integrity (20%): matches {AVD} fingerprint, no regression from {MC}; if {VDA} exists, also evaluates whether {RW} addressed top drift offenders flagged in Step 13d.5 4. Narrative Quality (25%): coherent, authentic, well-paced 5. Technical Quality (15%): grammar, punctuation, formatting f. Character Arc Validation (if {ARC} exists): 1. Checks expected_state for chapter's characters 2. Compares against actual portrayal in {D} 3. DEVIATED = potential blocker (CRITICAL) or warning (minor drift) 4. Updates {ARC} actual_state and alignment fields g. Thematic Density Check (if {TKW} exists): 1. Compares keyword density between {MC} and {D} 2. Drop < 10%: acceptable 3. Drop 10-25%: warning 4. Drop > 25%: blocker (thematic drift) h. Checks for blockers (automatic REJECT regardless of score): 1. {KB} violation (detected by {CC}, but verified) 2. Voice regression (score dropped from original) 3. Incomplete {CTPL} (steps skipped without idempotency pass) 4. {AVD} constraint violation 5. Repetitive foreshadowing (same warning in multiple chapters) 6. Fact re-establishment (re-explaining established traits) 7. Motif overuse (same quirk in consecutive chapters) 8. Exposition dump (4+ sentences pure explanation without grounding) 9. Arc deviation (CRITICAL severity) 10. Thematic drift (>25% drop) 11. Voice marker floor violation i. Calculates Weighted Supervisor Score {SSCORE}. j. Decision: 1. Hard floor check: if {SSCORE} < {CAL}.supervisor_floor (default 70), mandatory {ESCALATE} regardless of cycle. Rescue Mode is NOT invoked below the floor. 2. {APPROVE}: {SSCORE} >= threshold AND zero blockers → proceed to Vision. 3. {REJECT}: {SSCORE} < threshold OR blocker present. a. Cycles 1-2: Send feedback to {CTP} for new plan. b. Cycle 3: Activate Rescue Mode, retry once. c. Cycle 3 rejection after Rescue: {ESCALATE} to {U}. k. Outputs: 1. Supervisor Review {SREV} to reviews/supervisor/chapter-[N]-review.md 2. Metrics JSON to metrics/chapter-[N]-supervisor-cycle-[C].json 18. The Vision Final Approver {VFA}: a. Determines current cycle: 1. cycle = {STAT}.current_cycle (authoritative source). 2. Fallback if {STAT} missing: parse cycle from {CTPL} filename (chapter-[N]-ctp-cycle[C].md). 3. Final fallback: cycle = count(metrics/chapter-[N]-vision-cycle-*.json) + 1. 4. Hard cap: max cycle = 2 for VFA review attempts. b. Inputs supervisor-approved {D}, {AVD}, {MEO} entry, {SREV}, {CAL}. c. If {SLINK} exists: also inputs {SKB} for series arc alignment. d. Evaluates 4 dimensions (pass/fail each): 1. AVD Thematic Alignment: embodies core themes, emotional register, narrative intent 2. Whole-Book Coherence: fits within complete arc, character trajectories, thematic progression 3. Series Arc Alignment (if {SER}): aligns with series-level arcs from {SKB} 4. Publishing Readiness: zero blocking issues, professional standards met e. Applies Author Satisfaction Test: does chapter deliver on original authorial intent from {AVD}? 1. Inputs: supervisor-approved {D}, {AVD} success criteria, {AVD} thematic core, {AVD} character development map. 2. For each of {AVD}'s 5 measurable success criteria, determines pass/fail by reading {D} for evidence of the criterion. 3. For {AVD}.thematic_core: confirms central themes are embodied in {D} (not merely mentioned). Failure mode: theme-as-label, not theme-as-pressure. 4. For {AVD}.character_development: confirms expected arc states from {ARC} (if present) for characters appearing in this chapter. 5. Decision: PASS if all 5 success criteria pass AND no theme-as-label finding AND no arc deviation. PARTIAL if 4/5 success criteria pass with no critical findings. FAIL otherwise. 6. PARTIAL escalates to {U} for explicit accept-with-known-shortfall or send-back. FAIL routes per Step 18f.2. 7. This is the only step in the pipeline where a human-judgment criterion (does it land?) is encoded; the other dimensions are mechanical checks. Author Satisfaction Test outputs MUST be reviewed by {U} before {VFA}.{APPROVE} is final, even if the automated test PASSes. f. Decision: 1. {APPROVE}: All dimensions pass → proceed to MEO Manager 2. {DECLINE}: Any dimension fails a. Cycle 1: Return to {CTP} with vision feedback b. Cycle 2 rejection: {ESCALATE} to {U} for manual revision g. Outputs: 1. Vision Review {VREV} to reviews/vision/chapter-[N]-review.md 2. Metrics JSON to metrics/chapter-[N]-vision-cycle-[C].json h. QA Lint (ALWAYS runs; verbosity controls only the report file, not the check): 1. Runs a lightweight QA pass over {D}: a. Confirms prose-only output (no agent commentary, no outlines, no tool logs) a.1 Confirms no Change Manifest, manifest delimiter, or JSON appears in {D}. b. Confirms no leaked placeholders (e.g., "{MC}", "{KB}", "{U}") in prose c. Confirms markdown hygiene (balanced fences, no broken headings) 2. Outputs QA report to reviews/qa/chapter-[N]-qa.md with {PASS}/{WARN}/{FAIL}. Report file written only when verbosity != minimal; the {PASS}/{WARN}/{FAIL} verdict itself is always recorded in {STAT}. 3. If {FAIL}: return to {CTP} with defect list (does not increment cycle). 19. The MEO Manager {MEOM}: a. Inputs {MC}, vision-approved {D}, {MEO}, {MEOJ}, {KB}, {DELTA}. b. Compares original vs approved to identify significant changes. c. Categorizes by type: 1. Character: traits, abilities, relationships 2. Plot: events, subplots 3. Timeline: dates, durations 4. World: rules, locations 5. Thematic: emphasis shifts d. Executes Delta Merge Protocol: 1. Acquires master KB write lock (status/kb-merge.lock; single-writer guarantee). If lock held by another chapter pipeline, queues and waits. {MEOM} MUST NOT proceed without the lock. 2. Reads {DELTA} entries for this chapter 3. Re-reads {KB} from disk under the lock (guards against stale-read races with another concurrently-completing chapter). 4. Validates no conflicts with {KB} 5. If valid: appends to {KB}, removes from {DELTA} 6. If conflict: flags for manual review 7. Releases write lock and emits status/kb-merge-log.json entry with timestamp, chapter_id, kb_hash_before, kb_hash_after. 5. Logs merge in status/kb-merge-log.json e. Analyzes downstream impacts (forward-looking only): 1. P1: Must fix - causes continuity break if ignored 2. P2: Should fix - improves coherence 3. P3: Could fix - polish opportunity f. Updates {MEOJ} with change_log entries: 1. timestamp, source_chapter, target_chapter 2. field, action (add/modify/remove), value, rationale g. If {SLINK} exists: flags {SCAN} updates needed. h. Outputs: 1. Propagation Log {PROP} (JSON) 2. Propagation Summary {PSUM} (Markdown) 20. The Back Strategist {BS}: a. Inputs {PROP} from {MEOM}. b. Identifies backward impacts affecting PREVIOUS chapters: 1. Retcons: new info contradicting earlier content 2. Timeline shifts: changed event timing 3. Character revelations: traits needing earlier hints 4. Object introductions: items needing setup 5. Relationship changes: dynamics evolved differently c. Categorizes by severity: 1. CRITICAL: creates plot hole if unfixed (reader notices) 2. IMPORTANT: weakens narrative coherence (missed foreshadowing) 3. OPTIONAL: polish opportunity (subtle improvement) d. Generates fix suggestions (does NOT auto-apply): 1. Chapter + anchor (preferred: scene_id from {SSF}; fallback: paragraph anchor text) 2. Issue description 3. Current text 4. Suggested replacement 5. Rationale e. If {ARC} exists: flags arc expectation updates for earlier chapters. f. Outputs Retroactive Fix List {RFL} to updates/retroactive-fixes/chapter-[N]-backward-impacts.md. 21. Status Consolidation: a. Updates Chapter Status {STAT} file with: 1. current_phase, current_cycle 2. enhancement_summary (started, completed, total_cycles, final_verdict) 3. diagnostics summary (scores, priority_areas) 4. reviews history (continuity, supervisor, vision) 5. artifacts paths (ctpl, rev, pol) 6. propagation summary (meo_updates, kb_flags, series_flags) b. Respects output_control.verbosity from {CAL}: 1. minimal: Only {REV}, Polished Chapter {POL}, {STAT} 2. standard: Core files + consolidated status 3. verbose: All files (legacy behavior) 22. Chapter Approval by {U}: a. {U} reviews vision-approved {D}: 1. Views changes via diff viewer 2. Reviews {RFL} for backward impacts b. Decision: 1. {APPROVE}: Chapter becomes Approved Chapter {AC} a. {AC} written to chapters/revised/ as prose only. b. Writes version snapshot to versions/: a. versions/chapter-[N]/approved-cycle-[C]-[YYYYMMDD-HHMM].md (copy of {AC}) b. versions/chapter-[N]/approved-cycle-[C]-manifest.json (inputs + review paths) c. Continue to next chapter 2. {REJECT}: Sends {D} with comments to {CTP} for reprocessing 3. {U} may approve specific {RFL} entries from {BS}; approval triggers Backward Fix Application (Step 22.5). STEP 22.5: Backward Fix Application {BFA} (triggered by {U} approval of {RFL} entries): a. Inputs target chapter {AC}, approved {RFL} entries, {KB}, {SSF}, {CREG}. b. Targeted patch only: applies the suggested replacement at the specified anchor (scene_id preferred). Does NOT re-run a full chapter cycle. c. Validation pass (lighter than full cycle): 1. {OV}: confirms patch produced a meaningful, anchored change. 2. {CC}: validate-only mode against {KB}, {SSF}, {CREG}; no threshold gate, but CRITICAL findings block. 3. QA Lint: same prose hygiene check as Step 18h. d. Decision: 1. {PASS}: writes patched {AC} as new version snapshot (versions/chapter-[N]/bfa-[YYYYMMDD-HHMM].md), updates {STAT} with bfa entry. 2. {FAIL}: reverts patch and surfaces failure to {U} with diagnostic. e. Outputs BFA log to updates/retroactive-fixes/chapter-[N]-bfa-log.json. END [PER-CHAPTER ENHANCEMENT LOOP] ================================================================================ PHASE 6: POST-ENHANCEMENT REVIEW ================================================================================ 23. Full Review {FR}: a. After all chapters processed, {U} may initiate full manuscript review. b. Discovery Phase: 1. Globs original/, revised/, polished/ directories 2. Verifies each original has revised version 3. Reports inventory c. Combination Phase: 1. Combines all {AC} into manuscript-revised.md 2. If polish exists: creates manuscript-polished.md 3. Creates manuscript-final.md (best available per chapter) d. Comparative Analysis: 1. Character consistency across chapters 2. Plot coherence and thread resolution 3. Timeline accuracy 4. Structure quality 5. Before/after diagnostic comparison e. Generates full-review-report.md with health scores and improvement metrics. f. {U} decision: 1. {APPROVE}: Accept all changes, mark ready for export 2. {REFINE}: Return specific chapters to pipeline 3. {REVIEW}: Request detailed breakdown of specific changes 4. {POLISH}: Proceed to polish phase ================================================================================ PHASE 7: POLISH PHASE (OPTIONAL) ================================================================================ START [OPTIONAL POLISH PHASE] 24. {U} initiates polish for approved chapters. a. Selection modes: 1. Single chapter: polish chapter N 2. Range: polish chapters 1-12 b. Gate check: User-approved {AC} must exist for each chapter, AND the corresponding {VREV} must record {APPROVE}. Vision approval alone is not sufficient (the user-approval gate at Step 22 is the final author check). 25. The Unified Polisher {UP}: a. Inputs user-approved {AC}, {AVD}, {CAL}. b. Mode selection: 1. Technical Mode (default): grammar, flow, clarity, consistency 2. Creative Mode: artistic suggestions (imagery, rhythm, metaphors) c. Technical Mode Execution: 1. Validates user-approved {AC} exists and the corresponding {VREV} records {APPROVE}. 2. Loads {AVD} voice fingerprint 3. Applies Polish Guidelines: a. Clarity: simplify complex sentences within {AVD} tolerance b. Flow: refine transitions without structural changes c. Grammar: fix mechanical errors (punctuation, typography) d. Consistency: standardize formatting, terms, spellings 4. Constraints (NOT allowed): a. Structural changes b. Content additions or deletions c. Reordering paragraphs d. Adding sensory images not implied e. Removing motifs/symbols f. Changing character motivations 5. Calculates voice metrics (before/after) 6. If voice score < threshold: reverts problematic changes 7. Self-check: confirms zero content/meaning changes 8. Outputs Polished Chapter {POL} d. Creative Mode Execution: 1. Reads {AC} in "beauty mode" for elevation opportunities: a. Flat descriptions → more evocative b. Predictable word choices → surprising c. Missed sensory opportunities d. Dialogue → sharper, more distinctive e. Metaphors → fresher f. Rhythm → more musical 2. For each section (~500-800 words): identifies up to 3 opportunities 3. Generates suggestions as DIFFS ONLY (never wholesale rewrites): a. Location (paragraph, line) b. Type: imagery/rhythm/word_choice/sensory/metaphor/dialogue c. Risk: SAFE/MODERATE/ADVENTUROUS d. Current text, Suggested text, Rationale 4. Maximum 10 suggestions per chapter 5. Does NOT auto-apply (requires {U} review) 6. Outputs Creative Suggestions {CSUG} to polish-reports/chapter-[N]-creative-suggestions.md 7. {U} may approve specific {CSUG} entries; approval triggers Creative Patch Application (Step 25.5). e. Creative Latitude: 1. ENCOURAGED: fresh metaphors, unexpected verbs, sensory depth, dialogue subtext, rhythm variation 2. PROHIBITED: plot changes, character behavior changes, new information, tone shifts, anachronisms, voice departures f. Output based on verbosity: 1. minimal: {POL} only 2. standard: {POL} + {STAT} update 3. verbose: {POL} + polish-report.md + polish-metrics.json STEP 25.5: Creative Patch Application {CPA} (triggered by {U} approval of {CSUG} entries): a. Inputs polished chapter {POL} (or {AC} if no prior polish), approved {CSUG} diffs. b. Targeted diff application only: applies suggested text replacements at specified locations. c. Validation pass: 1. {OV}: confirms diffs produced anchored changes within polish constraints (no structural shift). 2. QA Lint: prose hygiene check. 3. Voice metric check: confirms voice score did not drop below {CAL}.voice_threshold. d. Decision: 1. {PASS}: writes updated {POL} as new version snapshot, updates {STAT} with cpa entry. 2. {FAIL}: reverts diffs and surfaces failure to {U}. e. CPA does NOT invoke {SUP} or {VFA} (creative tier is gated by direct user approval). f. Outputs CPA log to polish-reports/chapter-[N]-cpa-log.json. END [OPTIONAL POLISH PHASE] ================================================================================ PHASE 8: FINAL ASSEMBLY ================================================================================ 26. Final Manuscript Assembly {FMA}: a. Combines {POL} files (or {AC} if no polish) into final manuscript. b. Naming: 1. manuscript-revised.md: all revised chapters 2. manuscript-polished.md: all polished chapters (if polish ran) 3. manuscript-final.md: best available version per chapter c. If {SLINK} exists: prompts {U} to run series-canon update. d. Writes version snapshots to versions/: 1. versions/manuscript-revised-[YYYYMMDD-HHMM].md 2. versions/manuscript-polished-[YYYYMMDD-HHMM].md (if polish ran) 3. versions/manuscript-final-[YYYYMMDD-HHMM].md ================================================================================ PHASE 9: SERIES CANON UPDATE (SERIES ONLY) ================================================================================ 27. The Series Canon Updater {SCU}: a. Verifies all chapters have user-approved {AC} files and corresponding {VREV} records with {APPROVE}. b. Inputs all {AC}, {KB}, {ARC}, {DELTA}, {SLINK}. c. Reviews {DELTA} for new facts by category: 1. New characters (names, roles) 2. Timeline events (what happened) 3. Relationship changes (who to whom) 4. World state updates (rule changes) d. Merges to {SCAN}: 1. Reads {SKB} and book's {KB} 2. Identifies new entries (not marked [INHERITED]) 3. Adds new entries with "[Added in Book N]" tag 4. Validates no conflicts with existing entries 5. Writes updated series-kb.md 6. Computes updated scan_hash and writes it to versions/foundation-lock.json. 7. Writes series canon version record to versions/series-canon-version.json (content_hash, schema_version, created_at). e. Updates {STL} with new events in chronological order. f. Prompts for Locked Elements candidates: 1. Character deaths 2. Major revelations 3. World rules established g. If {U} confirms: updates locked-elements.json. h. Updates book-registry.json: 1. status: "canonical" 2. completion_date: ISO-8601 3. chapter_count: X i. Reports: 1. Book ID 2. Characters added 3. Timeline events added 4. Elements locked 5. Series state: "Ready for Book N+1" ================================================================================ ESCALATION & ERROR HANDLING ================================================================================ ESCALATION TRIGGERS: 1. {CC} cycle 3 failure (Continuity Checker) 2. {CC} hard floor failure: {CSCORE} < {CAL}.continuity_floor 3. {SUP} cycle 3 rejection after Rescue Mode (Chapter Supervisor) 4. {SUP} hard floor failure: {SSCORE} < {CAL}.supervisor_floor 5. {VFA} cycle 2 decline (Vision Final Approver) 6. {OV} Silent Write Failure after 2 retries (Output Validator) 7. {SCAN} inaccessible (Series Canon) 8. Foundation lock mismatch or unresolved lock reconciliation after {U} declines regeneration/rollback ESCALATION PACKET {EPK} (REQUIRED OUTPUT ON ANY ESCALATION): 1. Saved to escalations/chapter-[N]-escalation-packet.json 2. Contents: a. Trigger + cycle + timestamp b. Pinned hashes (avd_hash, kb_hash, cal_hash, scan_hash if series) c. Inputs: {MC} hash, {D} hash, {CTPL}, relevant slices of {MEOJ} d. Violations: top 5 blocking findings from {CC}/{SUP}/{VFA}/{SCC} e. Minimal diffs: excerpted before/after around each violation (anchors) f. Recommended next actions (manual fix vs override vs skip) OVERRIDE FORMAT {OVR} (IF {U} CHOOSES "ADD OVERRIDE"): 1. Saved to overrides/chapter-[N]-override.json 2. Fields (minimum): a. scope: continuity | series | voice | rule b. rule_id (or anchor) being overridden c. allowed_severity: MINOR | MAJOR (CRITICAL not allowed for locked elements) d. rationale (why deviation is intentional) e. approved_by (name/initials) f. expires: chapter_only | until_book_end | permanent RESUME PROTOCOL (DETERMINISTIC): 1. On resume, agents load {OVR} (if present) BEFORE evaluation. 2. Overrides can downgrade findings (e.g., MAJOR→NOTE) but cannot erase locked-element CRITICALs. 3. Resume writes status/resume-log.md with: a. what was overridden b. what was rerun / skipped via idempotency c. new outputs + hashes ESCALATION OPTIONS FOR {U}: 1. Manual fix: {U} edits directly 2. Add override: Document intentional deviation (writes {OVR}) 3. Skip chapter: Proceed without enhancement 4. Abort pipeline: Stop and review TIMEOUT PROTOCOL: 1. Soft limit: 60 minutes (checkpoint progress) 2. Hard limit: 120 minutes (save state, pause) 3. State persisted to status/ for resume IDEMPOTENCY PROTOCOL: 1. Before executing any step: run success test 2. If pass: skip step, log as "already complete" 3. Ensures safe resume after interruption ================================================================================ CONCURRENCY & PLAN INVALIDATION ================================================================================ 1. Foundation hash change during in-flight chapter cycle: a. The active chapter saves checkpoint to status/chapter-[N]-checkpoint.json (validates against checkpoint.schema.json): 1. last_completed_step: phase + step id 2. pinned_input_hashes_at_start: snapshot from {LOCK} when cycle began 3. artifacts_in_flight: paths + partial content hashes 4. checkpoint_reason: "foundation_hash_change" | "timeout" | "user_pause" b. Cycle pauses; surface notification to {U} naming the changed artifact ({AVD}, {KB}, {CAL}, or {SCAN}) and the affected chapters. c. {U} chooses one of: 1. REGENERATE: discards in-flight work past last validation gate; rebuilds {CTPL} against new foundation; restarts cycle. 2. DEFER: continues current cycle ignoring the change until cycle completes; subsequent cycle picks up the change. Recorded as {OVR} with scope=continuity, expires=chapter_only. 3. ROLLBACK: reverts the changed foundation artifact to its previous content hash from versions/. Affected chapter resumes against the prior foundation. 2. {KB} delta merges from {MEOM} during in-flight cycles: a. {DELTA} entries are queued (not merged immediately) while any chapter cycle is active. b. Merge runs after the affected chapter completes its cycle, after the chapter pauses (case 1 above), or on explicit {U} request. c. Queued merges logged to status/delta-merge-queue.json. 3. Concurrent chapter cycles: a. Multiple chapters MAY be in flight simultaneously (e.g., {U} works chapter 7 while chapter 5 finishes). b. Each chapter cycle pins its own snapshot of {LOCK} at start (recorded in {SSNAP} and checkpoint). c. Foundation changes affect only chapters that have not yet started their next cycle; in-flight chapters complete against their pinned snapshot unless {U} chooses REGENERATE. ================================================================================ ARTIFACT RETENTION ================================================================================ 1. versions/ files are append-only and persist for the project's lifetime. 2. plans/, reviews/, metrics/ files older than 90 days from the current chapter cycle MAY be archived to versions/archive/ at {U} discretion. Default behavior is no automatic archival. 3. status/checkpoint files are deleted after successful cycle completion. 4. {DELTA} entries are deleted after successful merge (per Step 19d.3). 5. chapters/revised/.working/ files are kept for the duration of the active cycle and pruned on cycle completion (the canonical {AC} lives in chapters/revised/). ================================================================================ ================================================================================ ARTIFACT SCHEMA CONTRACTS & VERSIONING ================================================================================ All JSON artifacts MUST include schema_version and MUST be validated against schemas/ before use. 1. Schema Files (in schemas/): a. meoj.schema.json ({MEOJ}) a.1 mman.schema.json ({MMAN}, manuscript manifest) b. kb.schema.json ({KB} JSON if used; {KB} markdown remains human-readable) c. ssf.schema.json ({SSF}) d. stat.schema.json ({STAT}) e. override.schema.json ({OVR}) f. escalation.schema.json ({EPK}) g. creg.schema.json ({CREG}) h. ssnap.schema.json ({SSNAP}) i. change-manifest.schema.json ({CMAN}) j. supervisor-metrics.schema.json (Step 17k.2 metrics output) k. vision-metrics.schema.json (Step 18g.2 metrics output) l. polish-metrics.schema.json (Step 25 polish-metrics output) m. chapter-map.schema.json ({CMAP}) n. calibration.schema.json ({CAL}) o. theme-keywords.schema.json ({TKW}) p. arc-ledger.schema.json ({ARC}) q. rules.schema.json ({RULES}) r. propagation.schema.json ({PROP}) s. version-record.schema.json (all files under versions/ ending in -version.json, plus foundation-lock.json) t. book-registry.schema.json u. checkpoint.schema.json (status/chapter-[N]-checkpoint.json; see Concurrency section) v. manuscript-selection.schema.json (status/manuscript-selection.json) w. synthetic-origin-likelihood.schema.json ({SOL}) x. voice-drift-analysis.schema.json ({VDA}) y. kb-merge-log.schema.json (status/kb-merge-log.json) z. delta-merge-queue.schema.json (status/delta-merge-queue.json) aa. bfa-log.schema.json (updates/retroactive-fixes/chapter-[N]-bfa-log.json) ab. cpa-log.schema.json (polish-reports/chapter-[N]-cpa-log.json) ac. approval-manifest.schema.json (versions/chapter-[N]/approved-cycle-[C]-manifest.json) ad. lock-reconcile.schema.json (status/lock-reconcile.json) 1.1 JSON artifact policy: a. Every JSON artifact MUST have exactly one schema listed above. b. Logs that need schema validation MUST be JSON, not freeform text. c. If an artifact is intentionally Markdown or plain text, it is exempt from JSON schema validation but must not be described as a JSON artifact elsewhere. 2. Version Pins: a. versions/avd-version.json and versions/kb-version.json store content_hash + created_at for {AVD} and {KB}. b. versions/calibration-version.json stores content_hash + created_at for {CAL}. c. versions/series-canon-version.json stores content_hash + created_at for {SCAN}, if series mode is active. d. versions/chapter-[N]/ssf-baseline-version.json and versions/chapter-[N]/ssf-cycle-[C]-version.json store content_hash + created_at for chapter scene-state files. e. versions/foundation-lock.json is the single source of truth for pinned foundation hashes used in {CTPL}/{MEOJ}/{SSNAP}. f. If any pinned foundation artifact changes, downstream plans MUST be regenerated per Concurrency & Plan Invalidation section (prevents parallel drift). g. Chapter-local ssf_hash is pinned in {SSNAP} and {CTPL}; if it changes, only the affected chapter plan MUST be regenerated. PATH VARIABLES REFERENCE ================================================================================ Naming convention: actual chapter filenames use zero-padded two-digit numbering (chapter-07.md, chapter-12a.md). Path templates below use [N] as a placeholder for any chapter identifier including interludes. [C] is the cycle integer (1, 2, or 3). | Variable | Full Name | Path | |----------|--------------------------|---------------------------------------------------| | {MMAN} | Manuscript Manifest | narrative-project/manifest.json | | {MSRC} | Manuscript Source Record | narrative-project/status/manuscript-selection.json | | {CMAP} | Chapter Map | narrative-project/status/chapter-map.json | | {FREG} | Foreshadowing Registry | narrative-project/kb/foreshadowing-registry.json | | {OVR} | Override File | narrative-project/overrides/chapter-[N]-override.json | | {EPK} | Escalation Packet | narrative-project/escalations/chapter-[N]-escalation-packet.json | | {ASCH} | Schema Registry | narrative-project/schemas/ | | {LOCK} | Foundation Lock | narrative-project/versions/foundation-lock.json | | {UVS} | User Voice Sample | narrative-project/avd/voice-sample/user-voice.md | | {AVD} | Authorial Vision Doc | narrative-project/avd/authorial-vision.md | | {KB} | Knowledge Base | narrative-project/kb/knowledge-base.md | | {MEO} | Master Enhancement Outline| narrative-project/meo/master-outline.md | | {MEOJ} | MEO JSON | narrative-project/meo/meo.json | | {MC} | Manuscript Chapter | narrative-project/chapters/original/chapter-[N].md| | {REV} | Revised Chapter | narrative-project/chapters/revised/chapter-[N].md | | {POL} | Polished Chapter | narrative-project/chapters/polished/chapter-[N].md| | {CSUG} | Creative Suggestions | narrative-project/polish-reports/chapter-[N]-creative-suggestions.md | | {CTPL} | Chapter Tactical Plan | narrative-project/plans/chapter-[N]-ctp-cycle[C].md| | {STAT} | Chapter Status | narrative-project/status/chapter-[N]-status.json | | {CAL} | Calibration Profile | narrative-project/calibration.json | | {CDIAG} | Chapter Diagnostic | narrative-project/diagnostics/chapter-[N]-diagnostic.json| | {MDIAG} | Manuscript Diagnostic | narrative-project/diagnostics/manuscript-diagnostic.json| | {DELTA} | KB Delta | narrative-project/kb/kb.delta | | {CMAN} | Change Manifest | narrative-project/metrics/chapter-[N]-change-manifest-cycle-[C].json | | {VDA} | Voice Drift Analysis | narrative-project/diagnostics/voice-drift-analysis.json | | {SOL} | Synthetic Origin Likelihood | narrative-project/diagnostics/synthetic-origin-likelihood.json | | {SSF} | Scene State File | narrative-project/kb/scene-states/chapter-[N]-scenes.json| | {CREG} | Constraint Registry | narrative-project/kb/scene-states/chapter-[N]-constraints.json| | {SSNAP} | Status Snapshot | narrative-project/status/chapter-[N]-snapshot.json| | {ARC} | Character Arc Ledger | narrative-project/kb/arc-ledger.json | | {RULES} | World Rules | narrative-project/kb/rules.json | | {TKW} | Theme Keywords | narrative-project/avd/theme-keywords.json | | {GOLD} | Gold Excerpts | narrative-project/avd/gold-excerpts/ | | {SLINK} | Series Link | narrative-project/series-link.json | | {SCAN} | Series Canon | ../series-canon/ | | {SKB} | Series Knowledge Base | ../series-canon/series-kb.md | | {STL} | Series Timeline | ../series-canon/series-timeline.md | | {SAVD} | Series Voice Baseline | ../series-canon/series-avd.md | | {PROP} | Propagation Log | narrative-project/updates/chapter-[N]-propagation.json| | {PSUM} | Propagation Summary | narrative-project/updates/chapter-[N]-propagation-summary.md| | {RFL} | Retroactive Fix List | narrative-project/updates/retroactive-fixes/chapter-[N]-backward-impacts.md| | {BFALOG} | BFA Log | narrative-project/updates/retroactive-fixes/chapter-[N]-bfa-log.json | | {CPALOG} | CPA Log | narrative-project/polish-reports/chapter-[N]-cpa-log.json | | {SREV} | Supervisor Review | narrative-project/reviews/supervisor/chapter-[N]-review.md| | {VREV} | Vision Review | narrative-project/reviews/vision/chapter-[N]-review.md| ================================================================================ AGENT REFERENCE ================================================================================ | Agent | Abbrev | Model | Purpose | |---------------------------|---------|--------|--------------------------------------------| | Initializer | {INIT} | - | Project setup, chapter extraction | | Diagnostic Analyzer | {DA} | sonnet | Pacing, sensory, emotional, foreshadowing | | Calibrator | {CALR} | - | M.A.P. calibration game | | Manuscript Visionary | {MV} | opus | AVD, theme keywords, arc ledger | | Knowledge Base Builder | {KBB} | sonnet | KB with 9 sections | | Gold Excerpt Extractor | {GEE} | opus | Voice conditioning excerpts | | Literary Master Planner | {LMP} | opus | MEO creation | | World Rules Generator | {WRG} | sonnet | World rules extraction | | Series Continuity Checker | {SCC} | sonnet | Series-canon validation | | Scene Parser | {SP} | opus | Scene state extraction | | Chapter Tactical Planner | {CTP} | opus | Tactical plan creation (4-pass) | | Chapter Rewriter | {RW} | opus | Tactical plan execution | | Output Validator | {OV} | - | Silent write failure detection | | QA Linter | {QA} | sonnet | Prose-only + formatting sanity check | | Continuity Checker | {CC} | sonnet | KB/series/rules validation | | Chapter Supervisor | {SUP} | opus | 5-dimension quality gate | | Vision Final Approver | {VFA} | opus | Thematic/voice final gate | | MEO Manager | {MEOM} | opus | Ripple effects, MEO updates | | Back Strategist | {BS} | opus | Backward impact analysis | | Unified Polisher | {UP} | opus | Technical/creative polish | | Series Canon Updater | {SCU} | - | Merge changes to series canon | ================================================================================ ABBREVIATION INDEX ================================================================================ Defined in order of first appearance: | Abbrev | Full Name | Defined In | |---------|------------------------------|------------| | {U} | User | Step 1 | | {M} | Manuscript | Step 1 | | {REQ} | Authorial Requirements | Step 1a | | {SER} | Series | Step 1b | | {VSRC} | Voice Source | Step 1c | | {UVS} | User Voice Sample | Step 1d | | {INIT} | Initializer | Step 2 | | {PROJ} | Project Directory Structure | Step 2b | | {MSRC} | Manuscript Source | Step 2a | | {MMAN} | Manuscript Manifest | Step 2a.2 | | {CMAP} | Chapter Map | Step 2c.5 | | {CMV} | Chapter Map Validator | Step 2c.6 | | {CMAPO} | Chapter Map Override | Step 2c.7 | | {ASCH} | Schema Registry | Step 2h | | {MC} | Manuscript Chapters | Step 2c | | {SLINK} | Series Link | Step 2f.1 | | {SCAN} | Series Canon | Step 2f.2 | | {DA} | Diagnostic Analyzer | Step 3 | | {CDIAG} | Chapter Diagnostic | Step 3c | | {MDIAG} | Manuscript Diagnostic | Step 3d | | {FREG} | Foreshadowing Registry | Step 3e | | {CALR} | Calibrator | Step 4 | | {CAL} | Calibration Profile | Step 4f | | {MV} | Manuscript Visionary | Step 5 | | {VFP} | Voice Fingerprint | Step 5c | | {MVFP} | Manuscript Voice Fingerprint | Step 5e | | {AVD} | Authorial Vision Document | Step 5j.1 | | {TKW} | Theme Keywords | Step 5j.2 | | {ARC} | Character Arc Ledger | Step 5j.3 | | {KBB} | Knowledge Base Builder | Step 6 | | {KB} | Knowledge Base | Step 6d | | {GEE} | Gold Excerpt Extractor | Step 7 | | {GOLD} | Gold Excerpts | Step 7g | | {LMP} | Literary Master Planner | Step 8 | | {LOCK} | Foundation Lock | Step 2i | | {MEOJ} | MEO JSON | Step 8f.1 | | {MEO} | Master Enhancement Outline | Step 8f.2 | | {WRG} | World Rules Generator | Step 9 | | {RULES} | World Rules | Step 9d | | {SCC} | Series Continuity Checker | Step 10 | | {SKB} | Series Knowledge Base | Step 10c.1 | | {STL} | Series Timeline | Step 10c.2 | | {SAVD} | Series Voice Baseline | Step 10c.3 | | {SP} | Scene Parser | Step 11 | | {SSF} | Scene State File | Step 11f | | {SSNAP} | Status Snapshot | Step 11.5 | | {CREG} | Constraint Registry | Step 12d.4 | | {CTP} | Chapter Tactical Planner | Step 12 | | {PF} | Plot Facts | Step 12d.1 | | {EN} | Emotional Non-Negotiables | Step 12d.2 | | {MM} | Motifs & Mechanics | Step 12d.3 | | {CTPL} | Chapter Tactical Plan | Step 12k | | {RW} | Chapter Rewriter | Step 13 | | {REV} | Revised Chapter | Step 13b | | {DELTA} | KB Delta | Step 13e.7 | | {D} | Draft Prose | Step 13i | | {CMAN} | Change Manifest | Step 13i | | {VDA} | Voice Drift Analysis | Step 5j.4 | | {SOL} | Synthetic Origin Likelihood | Step 3f | | {BFA} | Backward Fix Application | Step 22.5 | | {BFALOG}| BFA Log | Step 22.5f | | {OV} | Output Validator | Step 14 | | {CC} | Continuity Checker | Step 16 | | {CSCORE}| Continuity Score | Step 16k | | {SUP} | Chapter Supervisor | Step 17 | | {SSCORE}| Supervisor Score | Step 17i | | {SREV} | Supervisor Review | Step 17k.1 | | {VFA} | Vision Final Approver | Step 18 | | {VREV} | Vision Review | Step 18g.1 | | {MEOM} | MEO Manager | Step 19 | | {PROP} | Propagation Log | Step 19h.1 | | {PSUM} | Propagation Summary | Step 19h.2 | | {BS} | Back Strategist | Step 20 | | {RFL} | Retroactive Fix List | Step 20f | | {STAT} | Chapter Status | Step 21a | | {POL} | Polished Chapter | Step 21b.1 | | {AC} | Approved Chapter | Step 22b.1 | | {FR} | Full Review | Step 23 | | {UP} | Unified Polisher | Step 25 | | {CSUG} | Creative Suggestions | Step 25d.6 | | {CPA} | Creative Patch Application | Step 25.5 | | {CPALOG}| CPA Log | Step 25.5f | | {FMA} | Final Manuscript Assembly | Step 26 | | {SCU} | Series Canon Updater | Step 27 | | {OVR} | Override File | Escalation | | {EPK} | Escalation Packet | Escalation | ================================================================================ END OF PROCESS SPECIFICATION ================================================================================